Setting Up Tracked Prompts

Every scan asks AI engines a set of questions and analyses the answers. Out of the box those questions are generated for you from the project’s brand, industry and competitors. That is fine for a first scan and wrong for a serious programme, because the questions your buyers actually type are specific to your market and nobody can infer them from a domain name.

Tracked Prompts is where you take that over. It lives in the Monitor group of the sidebar, subtitled “Define the buyer-intent queries your brand is monitored against”.

Step 1: Understand what tracking a prompt changes

The explainer card at the top of the page states the rule, and it is a replacement rather than an addition: when a project has active tracked prompts, scans query the engines with exactly those. Leave the list empty (or switch every prompt off) and scans fall back to the auto-generated discovery prompts. There is no blended mode.

Write prompts the way your buyers actually ask, not the way you would phrase a keyword. Google’s own guidance on AI features makes the same point from the other side: these surfaces answer conversational, intent-bearing questions rather than matching terse strings, so “what’s the best project management tool for a remote design team” is a far more useful thing to monitor than “project management software”.

The card also carries a badge showing how many prompts are currently active, and a note about cost. The unit is the prompt, not the engine call: one credit buys one prompt, checked once, across every model you run. Ten active prompts is ten credits a scan whether one model answers them or all the models you are entitled to. So the prompt count is the only lever on cost: adding prompts adds usage in a straight line, and picking fewer models saves nothing while costing you coverage.

Step 2: Discover a starting set

You do not have to start from a blank page. The Discover prompts card runs an AI pass over your project that proposes the buyer-intent questions your market asks, clusters them into topics, and estimates monthly volume for each. Click Discover prompts and the work happens in a background job; a toast tells you it has started, or that a run is already in progress.

Discovery is limited to one run per project per hour, and that limit is a refusal rather than a quiet reuse: click again inside the window (even after a run has finished) and you get an error toast saying discovery already ran for this project in the past hour. The window is measured from the last run’s start, not from the top of the clock. Viewers cannot start one at all: the role bar is member and above, as with everything that spends AI credits.

When it finishes, two things change on the page. Topic chips appear under the card, each showing how many prompts fall in that cluster and their combined monthly volume. And new suggestions land in your prompt list inactive, which is deliberate: discovery proposes, you decide. Nothing starts consuming credits because a job ran overnight.

Volumes come from a real keyword-volume provider when one is configured for the deployment, and from the model’s own coarse estimate when it is not. Treat them as ordering information (the relative ranking is what you use) rather than as precise search counts.

Step 3: Add the prompts discovery will not think of

Under Add a prompt, type the question and press Enter or click Add prompt. The placeholder shows the register that works: best project management tool for startups, a real question in a buyer’s words, not a keyword.

Write them as a person would ask an assistant. “Best CRM for a five-person sales team” beats “CRM software” because it is what someone actually types into a chat box, and because the answer will name a shortlist you can be on or absent from. Cover the shapes that matter commercially: category questions (“best X for Y”), comparison questions (“X vs Y”), alternative questions (“alternatives to competitor”), and objection questions (“is X worth it for Z”).

If you re-add something you already track, the page refuses with “That prompt is already tracked” rather than creating a duplicate that would quietly double its cost.

Step 4: Activate only what you will act on

Each row in Your prompts has a toggle on the left. Active prompts run in scans; inactive ones sit in the list costing nothing. The row shows the prompt text, whether it is active, when it was created, and (for discovered prompts) a topic badge, a monthly volume figure and a priority badge.

The pencil icon edits the text in place (Enter saves, Escape cancels). The bin icon removes the prompt after a confirmation dialog that spells out the consequence: it will no longer be used in scans for this project.

Curate ruthlessly. A prompt you would not change your content over is a prompt spending a credit every scan to produce a row you scroll past.

Step 5: Know the active-prompt cap

Active prompts you may hold is a per-project cap: 3 on Free, 15 on Business, 100 on Agency. The figure is the same as the plan’s prompt allowance (one period’s scan on the free plan, one day’s on the paid plans), so a project can never store a prompt list its own scans would truncate.

Note per project, not per workspace: a three-domain Business workspace gets 15 active prompts on each of its projects. The Tracked prompts figure on the pricing page is the same number (prompts are standing, and a period’s allowance is what one scan can run), so there is no second, workspace-wide pool to keep straight. The cap is what refuses you at the point of clicking.

The cap is enforced when you add a prompt and again when you switch an inactive one on, and it answers with an explicit message naming your plan and cap: “The business plan allows 15 active prompts. Deactivate one, or upgrade to track more.” The check takes a lock and counts in the same transaction, so two people activating prompts at the same moment cannot both slip past it.

A scan runs every active prompt. The per-scan ceiling is the same number, which is the point: there is no window smaller than your active list for prompts to fall outside of, and nothing to reorder or plan around. If you want a prompt out of the next scan, deactivate it: that is the only control, and it is non-destructive, whereas deleting a prompt permanently discards the topic, volume and priority metadata that only a discovery run can write.

One case does produce truncation, and it is worth recognising rather than designing for: downgrade to Free while holding more active prompts than Free’s cap of 3. Nothing deactivates prompts when a plan drops, so the scan takes them in creation order, oldest first, and leaves the rest. It says so rather than hiding it: the job’s entry on Activity reads “N prompt(s) over the per-scan limit were not run”.

Step 6: Maintain the list

Tracked prompts are not a set-and-forget setting. Three habits keep them honest:

• Re-run discovery periodically. New suggestions arrive inactive, so a run is safe; you review the topic chips and pull in anything the market started asking that you were not watching.
• Watch the credit maths before you expand. Five more active prompts is five more credits every scan (daily on paid plans, weekly on free, plus anything you run by hand), however many engines you have on.
• Retire prompts that stopped teaching you anything. A question you have ranked first on for six months is confirmation, not information.

Deactivating everything is a legitimate state, not a broken one: the project simply returns to auto-generated discovery prompts, which is the right setting for a domain you are only watching (a competitor, say) rather than actively optimising.

Step 7: Where to go next

• Prompt monitoring strategy: how to choose a prompt set that stays useful over quarters, not weeks.
• Finding AEO queries: research methods for the questions discovery will not surface on its own.
• Multi-engine monitoring: where each of these prompts gets asked, and why enabling every engine costs no more than enabling one.
• Understanding Answer Engine Rankings: what the scans do with the prompts once you have chosen them.